Breadcrumb

  1. Inicio
  2. node
  3. Written Testimony of Joshua W. Mitchell Senior Economist, Welch Consulting

Written Testimony of Joshua W. Mitchell Senior Economist, Welch Consulting

Meeting of November 20, 2019 - EEOC Convenes Public Hearing on the Proposed Revision of the Employer Information Report (EEO-1)

Chair Dhillon and Commission Members:

It is an honor to appear before you today to discuss the important topic of collecting pay data as part of the EEO-1 Component 2 program. 

Before weighing the benefits and costs of continued Component 2 collection, it is helpful to review the elements of a pay equity analysis-the standard statistical framework for estimating gender and race/ethnicity pay differences.  A pay equity analysis may be undertaken in the context of litigation, initiated by an OFCCP audit (for federal contractors), or done proactively by self-auditing companies.  The analysis usually involves extracting employee-level microdata from a company's Human Resource Information System (HRIS).  The outcome variable is a compensation measure of interest that is usually expressed as a pay rate, such as an employee's full-time equivalent salary or quarterly bonus earned.  Other key data fields include gender and race/ethnicity demographics, a job category field that meaningfully groups together workers who are similarly situated, and other legitimate pay factors that reflect how that particular company sets pay.  Depending on the company, these pay factors might include the level of work being performed, an employee's time in their current role, overall company tenure, relevant outside work experience, educational attainment, certificates earned, performance review scores, etc. The microdata are typically analyzed with a multivariate regression model.  Output from a pay equity analysis reveals whether there is a statistically significant, unexplained gender or race/ethnicity pay disparity among similarly-situated workers, after accounting for legitimate pay factors.    

Potential Utility of Component 2

Turning to the EEO-1 Component 2 form, it is important to keep in mind the ultimate objective-what are the EEOC and OFCCP hoping to use these data for?  The most important objective is to enable the EEOC and OFCCP to better target enforcement resources. In order to meet this objective, it must be the case that an analysis of the Component 2 pay data can meaningfully distinguish between companies that are and are not complying with the Equal Pay Act and Title VII of the Civil Rights Act (and to a sufficient degree as to justify the additional burden placed on employers and agencies).  This in turn means that an analysis of Component 2 aggregate pay band data must correspond well with what a formal pay equity analysis of the underlying microdata would show. 

There are, however, at least six reasons why an analysis of aggregate pay band data is unlikely to yield similar results to a formal pay equity analysis.  First, the EEO job categories are very broad and unlikely to properly group workers together who are truly similarly situated.  A reporting hospital, for example, must categorize such diverse occupations as doctors, nurses, and attorneys in the same "Professional" EEO job category.   The courts have rejected broad job categories in Title VII compensation litigation and even the OFCCP in its Directive 2018-05 advocates developing Pay Analysis Groupings of similarly situated employees.   Second, the W-2 Box 1 annual taxable wage concept is intended for IRS accounting purposes, but is ill-suited for comparing employee pay rates because it conflates employer and employee decisions and ignores employment changes that affect compensation.  Employees who are promoted halfway through the year, for example, will have low annual taxable wages even if they are receiving the same rate of pay as other similarly-situated workers at the time of the snapshot.  An additional problem with Box 1 of the W-2 is that it excludes nontaxable compensation such as employee payments for health insurance premiums and contributions to 401(k)-type plans.  As a result, two workers could receive the exact same gross pay but appear to be paid differently according to Box 1 because one of the workers contributed more to a retirement account.  Third, the 12 pay band reporting-scheme which was created as part of the sampling design for the BLS Occupational Employment Statistics (OES) survey, is unlikely to reflect the compensation distribution of most companies' pay structures.[1]  Collapsing the data into pay bands can both exaggerate differences between employees who are paid just above and below an arbitrary cutoff, as well as hide considerable pay differences among employees in the same pay band.  Fourth, the grouping together of non-exempt and exempt workers and the imputation of 20 hours for part-time work and 40 hours for full-time work to all exempt employees is not consistent with many companies' employment practices.  Fifth, the Component 2 form doesn't collect information on any legitimate factors that could explain differences in employee pay.  This means an employee who is just starting his or her career is treated the same as another employee who has worked at the company for 30 years.  Sixth, the kinds of statistical testing available with aggregate pay band data, such as Mann-Whitney and interval regression, are unlikely to correspond well with the standard statistical testing used in employee-level multivariate regression analysis, and there is no obvious way to integrate the collected hours information into the testing of pay band representation differences. 

For all of the above reasons, attempts to detect discrimination with the Component 2 data will likely lead to high rates of false positives (Type I errors), which means many companies will be  flagged for potential violations despite fully complying with the law.  Similarly, analysis of Component 2 data will likely lead to high rates of false negatives (Type II errors), which means companies that are violating the law will not be flagged for further investigation. 

How high will these error rates be?  We, at Welch Consulting, simulated in 2016 the ability to detect discrimination using nationally representative data from the 2013 and 2014 Current Population Survey Earner Study.[2]  Across thirteen industry/EEO occupation categories, the key finding was that targeting enforcement based on the Component 2 form would do little better at detecting discrimination than selecting firms for audit completely at random. This held whether enforcement was based on statistical significance as indicated by the Mann-Whitney test or based on employer performance relative to other firms within the same industry.     

Earlier research by Abt Associates reinforces this conclusion.  In their study of OFCCP contractors, they compared findings from the EO survey, which collected summary pay data, to the results of completed OFCCP investigations of the same employers.  That is, they were able to test the predictive power of summary pay data using actual investigations where the findings were already known.  As described by the 2012 National Research Council panel, the principle conclusion was that the EO Survey "did not improve deployment of enforcement resources toward contractors most likely to be out of compliance and did not lead to greater self-awareness or encourage self-evaluations."[3]

Finally, it should be noted that the EEOC already receives aggregate pay band data on the EEO-4 form as part of the state and local government data collection program.  When interviewed about the use of this data by the same National Research Council panel, both the DOJ and EEOC indicated that the pay data was not useful.[4]     

Summarizing, both our simulation studies and EEOC's and OFCCP's prior experiences using aggregate pay data to guide enforcement efforts indicate that these data are no better than flipping a coin in detecting potentially non-compliant employers.

Burden of Component 2

The Paperwork Reduction Act also requires an estimate of respondent burden.  EEOC's initial estimate of the burden imposed on employers is likely understated for at least four reasons.  First, EEOC does not adequately account for the fact that the number of requested data elements had expanded from 140 to 3,360 per form and year of collected data.  Careful preparation of an EEO-1 form requires more than simply pushing a button.  It is not uncommon for a team of HR employees to be involved in the data verification process as well as an attorney, given that companies must certify that their submission is accurate. Second, the decision to use W-2 pay data significantly complicates the preparation of the form because HRIS and payroll systems are often not well integrated. IT professionals may have expertise in one data system but not the other, which further increases the number of people involved in preparing the submission. In my experience working with HRIS, payroll, and timekeeping data, it is not uncommon for these systems to fail to align perfectly on the first attempt.  Investigating mismatches requires additional time and attention. Third, as acknowledged in its September 12, 2019 Federal Register Notice, EEOC changed its longstanding practice of estimating the burden per form to estimating the burden per employer.[5] This change was inconsistent with guidance provided by GAO and OMB and resulted in a substantial reduction in the estimated hours required to complete Component 2.  Fourth, we have anecdotal evidence from clients who had difficulties preparing the form and uploading the completed form to the NORC portal.  In one instance, after several attempts on its own and spending several hours on the phone with NORC's support staff, the client asked that we assist in uploading the form for them. The time involved in resolving this issue alone would substantially raise any estimate of respondent burden.     

Conclusion

Given the lack of utility of Component 2 data for detecting discrimination and the fact that some additional burden is inevitably imposed on employers (and considerably more than was initially estimated), it would be difficult to recommend its continued collection under the terms set forth in the Paperwork Reduction Act.  While I am committed to ensuring pay equity, it is clear to me that Component 2 collection does not further this goal. 

 

Thank you for your time and I look forward to your questions.



[1] The OES pay bands also correspond to a pay rate concept that differs from W-2 Box 1 earnings. 

[2] EEO-1 Comments, U.S. Chamber of Commerce, April 1, 2016.  Exhibit 4, pp.110-146. 

[3] National Research Council. 2012.  Collecting Compensation Data from Employers.  Washington, D.C: The National Academies Press.  https://doi.org/10.17226/13496  p. 36.

[4] National Research Council. 2012.  Collecting Compensation Data from Employers.  Washington, D.C: The National Academies Press.  https://doi.org/10.17226/13496  pp. 19, 28.

[5] Agency Information Collection Activities: Existing Collection. Federal Register 84: 177.  pp. 48138-48142.